Inspect Command

The entrenar inspect command provides data and model inspection capabilities for SafeTensors, GGUF, Parquet, and CSV files.

Usage

entrenar inspect <INPUT> [OPTIONS]

Arguments

ArgumentDescription
<INPUT>Path to file to inspect (.safetensors, .gguf, .parquet, .csv)

Options

OptionDescription
--mode <MODE>Inspection mode: summary, outliers, distribution, schema (default: summary)
--columns <COLS>Specific columns to inspect (comma-separated)
--z-threshold <Z>Z-score threshold for outlier detection (default: 3.0)
-v, --verboseShow detailed tensor/column information

Inspection Modes

Summary Mode (default)

Display basic file statistics and metadata:

entrenar inspect model.safetensors

Output:

Inspecting: model.safetensors
  Mode: summary
Model Information:
  File size: 125.42 MB
  Parameters: 0.03B
  Tensors: 48

Outliers Mode

Detect statistical outliers using z-score:

entrenar inspect data.parquet --mode outliers --z-threshold 2.5

Output:

Inspecting: data.parquet
  Mode: outliers
Outlier Detection (z-threshold: 2.5):
  Column 'loss': 12 outliers found
  Column 'accuracy': 3 outliers found

Distribution Mode

Show distribution statistics for numeric columns:

entrenar inspect data.parquet --mode distribution

Schema Mode

Display schema information:

entrenar inspect data.parquet --mode schema

Supported Formats

FormatExtensionInformation Shown
SafeTensors.safetensorsTensor count, shapes, dtypes, total parameters
GGUF.ggufFile size, format version
Parquet.parquetFile size, column schema, row count
CSV.csvFile size, column names

Examples

Inspect SafeTensors Model

# Basic inspection
entrenar inspect model.safetensors

# Verbose output with tensor details
entrenar inspect model.safetensors -v

Output with -v:

Model Information:
  File size: 125.42 MB
  Parameters: 0.03B
  Tensors: 48

Tensor Details:
  model.embed_tokens.weight: [32000, 4096] (F32)
  model.layers.0.self_attn.q_proj.weight: [4096, 4096] (F32)
  model.layers.0.self_attn.k_proj.weight: [4096, 4096] (F32)
  ... and 45 more tensors

Inspect Training Data

# Summary of parquet file
entrenar inspect data/train.parquet

# Detect outliers in specific columns
entrenar inspect data/train.parquet --mode outliers --columns loss,accuracy

# Show distribution statistics
entrenar inspect data/train.parquet --mode distribution

Inspect GGUF Model

entrenar inspect model.gguf

Output:

GGUF Model Information:
  File size: 4.12 GB
  Format: GGUF (llama.cpp compatible)
  (Use llama.cpp for detailed GGUF inspection)

Programmatic Usage

#![allow(unused)]
fn main() {
use entrenar::config::{load_config, InspectMode};
use safetensors::SafeTensors;

// Load and inspect SafeTensors
let data = std::fs::read("model.safetensors")?;
let tensors = SafeTensors::deserialize(&data)?;

for name in tensors.names() {
    let tensor = tensors.tensor(name)?;
    println!("{}: {:?} ({:?})", name, tensor.shape(), tensor.dtype());
}
}

See Also